Concepts

How language models work, what they’re good at, and where they break.

0%

1. getting started
2. halfway
3. almost there
4. complete

explanationHow a language model workstutorialWatch a tiny agent work

tutorialWhat the model can seetutorialInstruction, context and formatexplanationWhere a model makes things uptutorialSame prompt, different answertutorialGetting a straight answerexplanationAssistant or agent?tutorialAnswers from documents the model never saw

tutorialShow one exampletutorialOutput the next step can useexplanationThe prompt before the conversationtutorialChange the prompt before you retrytutorialWhat every token costsexplanationWho decides the next step

explanationWhere memory comes from

Review: nothing due yet

About this course

Goals

• Explains how a language model produces text and where it fails
• Writes prompts that get reliable results
• Recognizes an agent, its tools and its degree of autonomy

Counts

16 lessons, 37 checkpoints

Builds on

Nothing; start here.

What AI models are and how they behave. Progress is kept in this browser; see your progress.

Lesson plan (16 lessons, 16 live)

#	Lesson	Status	Mode	Minutes	Covers	Serves	Exercise	After	Issue
How language models work
1	How a language model worksconcepts/how-models-work	live	explanation	20 estimate	how-models-work	explains-generationnames-failure-modes	judge
2	What the model can seeconcepts/context-window	live	tutorial	25 estimate	how-models-work	explains-generation	do	How a language model works
Prompting
3	Instruction, context and formatconcepts/prompt-anatomy	live	tutorial	25 estimate	prompting	structures-a-prompt	do	How a language model works
4	Show one exampleconcepts/one-example	live	tutorial	30 estimate	prompting	structures-a-prompt	do	Instruction, context and format
5	Output the next step can useconcepts/structured-output	live	tutorial	35 estimate	prompting	asks-for-structure	do	Instruction, context and format
6	The prompt before the conversationconcepts/system-prompt	live	explanation	15 estimate	prompting	structures-a-prompt	judge	Instruction, context and format
7	Change the prompt before you retryconcepts/iterating-on-a-prompt	live	tutorial	30 estimate	prompting	iterates-on-output	do	Output the next step can use
Capabilities and limits
8	Where a model makes things upconcepts/thin-path	live	explanation	20 estimate	limits	names-failure-modes	judge	How a language model works
9	Same prompt, different answerconcepts/same-prompt-twice	live	tutorial	30 estimate	limits	explains-generationnames-failure-modes	do	How a language model works
10	Getting a straight answerconcepts/straight-answer	live	tutorial	25 estimate	limits	names-failure-modes	do	Where a model makes things up
11	What every token costsconcepts/what-tokens-cost	live	tutorial	25 estimate	limits	explains-generation	do	What the model can see	#137
What an agent is
12	Watch a tiny agent workconcepts/agent-loop	live	tutorial	25 estimate	what-is-an-agent	names-the-loop	do	Change the prompt before you retry, Where a model makes things up	#140
13	Assistant or agent?concepts/assistant-or-agent	live	explanation	30 estimate	what-is-an-agent	tells-agent-from-assistant	judge	Watch a tiny agent work	#143
14	Who decides the next stepconcepts/autonomy-scale	live	explanation	25 estimate	what-is-an-agent	places-on-autonomy-scale	judge	Assistant or agent?	#146
Grounding and memory
15	Answers from documents the model never sawconcepts/retrieval	live	tutorial	20 estimate	grounding	explains-grounding	judge	Watch a tiny agent work	#149
16	Where memory comes fromconcepts/memory	live	explanation	25 estimate	grounding	explains-groundingexplains-generation	judge	Answers from documents the model never saw, What every token costs	#152